CN115859186A - Distributed optical fiber sensing event identification method and system based on Grarami angular field - Google Patents
Distributed optical fiber sensing event identification method and system based on Grarami angular field Download PDFInfo
- Publication number
- CN115859186A CN115859186A CN202310125487.5A CN202310125487A CN115859186A CN 115859186 A CN115859186 A CN 115859186A CN 202310125487 A CN202310125487 A CN 202310125487A CN 115859186 A CN115859186 A CN 115859186A
- Authority
- CN
- China
- Prior art keywords
- optical fiber
- event
- fiber sensing
- data
- distributed optical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention provides a distributed optical fiber sensing event identification method and a system based on a Grarami angular field, which relate to the technical field of data identification, wherein the method comprises the steps of acquiring optical fiber sensing signals acquired in a set scene and position information of distributed optical fiber sensors acquiring the optical fiber sensing signals in the set scene; extracting event data in the optical fiber sensing signal according to the sampling frequency when the optical fiber sensing signal is acquired and the position information of the distributed optical fiber sensor; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field; and inputting the acquired event image data into a student model trained by double migration and knowledge distillation to acquire a recognition result. The light weight capability of the recognition algorithm and the accuracy of the recognition algorithm are further improved.
Description
Technical Field
The disclosure relates to the technical field of data identification, in particular to a distributed optical fiber sensing event identification method and system based on a Grarami angular field.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
The optical fiber sensor realizes sensing monitoring by monitoring the change of physical quantity caused by measurement, wherein the distributed optical fiber sensor integrates sensing and transmission functions, can obtain the spatial and temporal multidimensional distribution information of the physical quantity, and has development prospect in the application fields of structure detection, leakage detection, transportation, perimeter security and protection, safety systems, optical fiber communication, environment measurement and the like.
In the initial research, distributed optical fiber sensing event recognition is performed by a method of machine learning and manual feature extraction, however, due to the problems of diversity and complexity of distributed optical fiber sensing application environments, long distance of signal monitoring, high sampling rate and the like, the method of machine learning has the disadvantages of manual feature extraction, incapability of processing mass data, and poor algorithm mobility in different environments. With the continuous development of deep learning, more and more researchers can realize the extraction and optimization of deep-level features of data by using a convolutional neural network to perform distributed optical fiber sensing event identification, so that the defects of a shallow machine learning method are overcome. However, in the process of identifying the distributed optical fiber sensing event by using the one-dimensional convolutional neural network (1D-CNN), when the method is applied to environments with inconsistent and complex data of the same type of events in different scenes, a serious overfitting problem exists, when the distributed sensing event is converted into an image, time correlation between signals cannot be kept in the aspect of characteristics, so that signal information is lost, and the problems of slow training of a large-scale deep network, complex training and low identification accuracy of a small-scale network exist through a common image classification network.
Disclosure of Invention
The distributed optical fiber sensing event identification method and system based on the Grarami angular field solve the problem that time correlation among signals cannot be reserved when event data are converted into images through a GAF (Grarami angular field) method, and migrate dark knowledge of a large-scale neural network into a user-defined light-weight neural network through a double-migration learning knowledge distillation method to ensure the accuracy of event identification.
According to some embodiments, the following technical scheme is adopted in the disclosure:
the distributed optical fiber sensing event identification method based on the Grarami angular field comprises the following steps:
acquiring optical fiber sensing signals acquired in a set scene and position information of a distributed optical fiber sensor acquiring the optical fiber sensing signals in the set scene;
extracting event data in the optical fiber sensing signal according to the sampling frequency when the optical fiber sensing signal is acquired and the position information of the distributed optical fiber sensor; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
and inputting the acquired event image data into a student model trained by double migration and knowledge distillation to acquire a recognition result.
According to some embodiments, the following technical scheme is adopted in the disclosure:
distributed optical fiber sensing event identification system based on the grassplot angle field comprises:
the data signal acquisition module is used for acquiring optical fiber sensing signals acquired in a set scene and position information of a distributed optical fiber sensor acquiring the optical fiber sensing signals in the set scene;
the event data conversion module is used for extracting event data in the optical fiber sensing signals according to sampling frequency when the optical fiber sensing signals are collected and position information of the distributed optical fiber sensors; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
and the event recognition module is used for inputting the acquired event image data into the student model which is trained by double migration and knowledge distillation to acquire a recognition result.
Compared with the prior art, the beneficial effect of this disclosure is:
the preprocessing method is simple, corresponding GAF event image data can be obtained by only carrying out a GAF (Granami angular field conversion) method on the collected optical fiber sensing signals, the calculation complexity and the processing time delay of the algorithm are reduced compared with the preprocessing process of the optical fiber sensing signals in other identification methods, such as wavelet packet denoising, empirical mode decomposition and the like, and meanwhile, the time domain correlation of the signals is stored in the GAF image according to the event characteristics.
By adopting the double-migration-network algorithm, the method not only enhances the capability of the recognition algorithm network in resisting model overfitting on the GAF image data set through a pre-training method, but also can improve the classification performance.
According to the method, the dark knowledge of the large model network can not be transferred to the lightweight network algorithm through the knowledge distillation method, the lightweight capability of the identification algorithm and the accuracy of the identification algorithm are further improved, and the method is greatly helpful for the practical application of distributed optical fiber sensing.
Drawings
The accompanying drawings, which are included to provide a further understanding of the disclosure, illustrate embodiments of the disclosure and together with the description serve to explain the disclosure and are not to limit the disclosure.
Fig. 1 is a flowchart of a distributed optical fiber sensing event identification method according to an embodiment of the present disclosure;
FIG. 2 is a structural and operational schematic diagram of a distributed optical fiber sensing system according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a process for converting event data into a corresponding GAF event signature by a Graham Angular Field (GAF) according to an embodiment of the present disclosure;
fig. 4 is a structural diagram corresponding to the student network model ICEblock in the embodiment of the present disclosure;
FIG. 5 is a block diagram of a CBAM attention module in a student network model ICEblock according to an embodiment of the present disclosure;
FIG. 6 is a block diagram of an ECANet attention module in the student network model ICEblock according to an embodiment of the disclosure;
FIG. 7 is a block diagram of a dual migration learning and knowledge distillation implementation of an embodiment of the disclosure.
Detailed Description
The present disclosure is further described with reference to the following drawings and examples.
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
Example 1
An embodiment of the present disclosure provides a distributed optical fiber sensing event identification method based on a glatiramer angular field, as shown in fig. 1, including the following steps:
the method comprises the following steps: acquiring optical fiber sensing signals acquired in a set scene and position information of a distributed optical fiber sensor acquiring the optical fiber sensing signals in the set scene;
step two: extracting event data in the optical fiber sensing signal according to the sampling frequency when the optical fiber sensing signal is acquired and the position information of the distributed optical fiber sensor; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
step three: and inputting the acquired event image data into a student model trained by double migration and knowledge distillation to acquire a recognition result.
Specifically, as an embodiment, the method for identifying the distributed optical fiber sensing event based on the Grammite Angular Field (GAF) and the light weight is provided, and the specific technical scheme is as follows:
step 1: the distributed optical fiber sensing is arranged in different application scenes, data acquisition is carried out according to the different application scenes, signal events are acquired at the positions where the distributed optical fiber sensing occurs, acquired one-dimensional signal data are obtained, and the corresponding data style is
。
And 2, step: and (3) uniformly dividing the signals acquired in the step (1) into data in 1S according to the sampling frequency during signal acquisition, and converting the event time sequence data in 1S into GAF images corresponding to different events in 1S through a Granami angular field. The size of the obtained event data set is adjusted and the event data set is divided into a training set, a verification set and a test set according to the proportion, and the proportion corresponding to the three data sets is as follows: 8:1:1.
And step 3: two different deep convolutional neural networks are designed, one is a teacher network model, and the other is a student network model. The teacher network model is pre-trained on the ImageNet dataset and the optimal training weights are saved. The student network model is pre-trained on the MNIST public data set and the best pre-selected weights are saved.
And 4, step 4: and (3) performing knowledge distillation training on the optimal training weight of the double migration networks, the teacher model and the student model in the step (3) on the GAF image event data set, and storing the training model parameters of the optimal student model.
And 5: and (4) carrying out unknown distributed optical fiber sensing event identification on the student model trained in the step (4).
Specifically, the process of uniformly dividing the signal acquired in step 1 into data in 1S according to the sampling frequency during signal acquisition, and converting the event time sequence data in 1S into GAF images corresponding to different events in 1S through a gray-scale angular field includes:
step 2-1: the event sequence collected in the step 1 is processed
Wherein->
The length of the time sequence is determined, and one-dimensional time sequence data is mapped to [ -1,1 ] by a Min-Max Scaler normalization method]Wherein, the formula corresponding to the corresponding normalization method is:
in the formula (I), the compound is shown in the specification,iindicated as a certain moment in time in the sequence of events,
is shown asiThe sensor data value corresponding to the instant in time->
Expressed as a sequence of events>
Is greater than or equal to>
Expressed as a sequence of events>
Is of the minimum value of (4), is greater than or equal to>
Expressed as obtained after normalizationiAnd sensing data values corresponding to the moments.
Step 2-2: converting the time sequence in the step 2-1 into a polar coordinate, namely regarding the numerical value as a cosine value of an included angle and regarding the timestamp as a radius, wherein the corresponding formula is as follows:
in the formula (2), the first and second groups,
expressed as obtained after normalizationiThe sensor data value corresponding to the instant in time>
Expressed as a normalized sequence of events, <' > based on the number of events>
Expressed as the calculated cosine value of the included angle, is greater than or equal to>
Is shown asiA time stamp in the sequence of moments, <' >>
Expressed as a calculated radius, N is a constant used as a polar space regularization factor, and the time dependence can be preserved by transforming the time series into polar coordinates.
In equation (2), N is a constant used as a polar space regularization factor, and the time dependency can be maintained by converting the time series into polar coordinates.
Step 2-3: after the numerical value is mapped to the polar coordinate, a corresponding Granami angular field matrix is obtained by calculating a trigonometric function value between two points, and the corresponding formula is as follows:
in the formula (I), the compound is shown in the specification,Gexpressed as a corresponding glatiramer angular field matrix calculated,
expressed as the calculated cosine value of the included angle, is greater than or equal to>
Expressed as calculating the value of a trigonometric function between two points.
Step 2-4: and scaling the elements in the Gerami angular field matrix to 0-255, thereby obtaining the event GAF two-dimensional image.
Specifically, in step 3, the specific structure of the student network model is as follows:
the student network module is mainly structurally of an inverted residual structure, and a space and channel attention CBAM module and a local cross-channel interaction strategy module ECANet are added into the inverted residual structure. Specifically, in order to highlight the lightweight effect of the model, a reversed residual structure in the MobileNet v2 is selected as a main structure of a student network module, in order to further improve the extraction capability of deep features of the event GAF diagram, multi-dimensional feature extraction and fusion are performed, and a space and channel attention CBAM module and a local cross-channel interaction strategy module ECANet are added into the reversed residual structure.
The student network model mainly comprises 2 Conv2d, 10 user-defined ICEblock and the last Linear layer Linear. The ICEblock is a main module of a student network model, data dimension lifting is carried out by using 1 x 1 convolution firstly in the ICEblock, then depth residual error convolution is carried out by using 3 x 3 convolution, channel attention mechanism and space attention mechanism fusion is carried out through a CBAM module, data dimension reduction operation is carried out through the 1 x 1 convolution, short link is used in a residual error structure only when the dimensions of input features and output features in the inverse residual error structure are the same, and ECANet is introduced into the inverse residual error structure for carrying out attention mechanism fusion again.
As an embodiment, in step 4, the process of performing knowledge distillation training on the event GAF image data set by using the optimal training weights of the dual migration network and the teacher model and the student model in step 3, and storing the training model parameters of the optimal student model includes:
step 4-1: loading the optimal teacher network for the migration training, inputting the event GAF image information into the model in the form of pixel matrix to obtain the probability distribution of corresponding different events, and successively dividing by the temperature
Smoothing is carried out, and then the soft label on the event data set is obtained by the softmax function>
Corresponding calculation formula (4), in which +>
The feature map of the last layer of the teacher network model.
In the formula (I), the compound is shown in the specification,
is represented as a characteristic diagram output by the last layer of the teacher model, T is a temperature parameter, n is the number of samples, and->
Represented as a soft label for the teacher model.
Step 4-2: loading the optimal student network model for the migration training, inputting event GAF image information into the model in the form of a pixel matrix to obtain probability distribution of different events, and successively dividing the probability distribution by the temperature
Smoothing is carried out, and then the soft label on the event data set is obtained by the softmax function>
Corresponding calculation formula (5), in which->
Is a feature map of the last layer of the student network model.
Wherein the content of the first and second substances,
for the feature map of the last layer of the student network model, be->
Is a soft label for a student network model.
Step 4-3: loading the optimal student network model for transfer training, inputting event GAF image information into the model in the form of pixel matrix to obtain probability distribution of different events, and directly obtaining hard labels of the student network model by softmax function
Corresponding calculation formula (6).
Wherein the content of the first and second substances,
a hard tag that is a student network model;
step 4-4: calculating the soft label of the teacher network model obtained in the step 4-1 and the soft label of the student network model obtained in the step 4-2 through a relative entropy loss function to obtain a soft loss function
The corresponding calculation formula is (7), in formula (7), B is the number of pictures to be processed in batch, and C is the number of category types of the event GAF image.
Wherein B is the total number of pictures in the batch, C is the total number of types of events, w is the number of pictures, v is the number of types,
soft tag, represented as a teacher model on an event data set, < > based on the teacher's context>
A soft tag, expressed as a student network model>
Expressed as a soft loss function of the teacher network model.
And 4-5: calculating the hard label of the student network model obtained in the step 4-3 on the GAF event data set and the real label of the event GAF image by using a cross entropy loss function to obtain a corresponding hard loss function value
The calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
expressed as a hard penalty function derived for the student network model>
Hard tags on event data sets, represented as student networks, in combination>
To representA real label of the sample picture; b is the total number of pictures in the batch, and C is the total number of categories of events.
And 4-6: the soft loss function obtained in the step 4-4 and the hard loss function obtained in the step 4-5 are subjected to a scaling coefficient
Adjusting the ratio of the two loss values results in a mixing loss function->
The corresponding calculation formula is (8), the resulting @>
The back propagation is the key for the teacher model to give the student model the dark knowledge and greatly improve the classification performance of the student model.
Wherein the content of the first and second substances,
for a mixing loss function, is>
Expressed as a hard loss function derived from a student network model,
is a proportionality factor->
Expressed as a soft loss function of the teacher network model.
And 4-7: and performing distillation training on the teacher model and the student models of the double migration networks on the GAF event data set, and storing the optimal training result of the student network model.
Example 2
In an embodiment of the present disclosure, taking long-distance pipeline safety monitoring as an example, a main flowchart of a distributed optical fiber sensing event identification method based on a Grammite Angular Field (GAF) and double-migration learning and knowledge distillation is shown in fig. 1, and mainly includes 5 steps:
step 1: and (5) signal acquisition. The method comprises the steps of laying distributed optical fiber sensors in an application scene, collecting different types of events by using distributed optical fiber sound wave/vibration sensing system hardware based on a phase-sensitive optical time domain reflectometer (phi-OTDR), dividing signal data into single-channel event data according to event space positioning and sampling frequency information, wherein the corresponding data pattern is
。
Step 2: and converting the optical fiber sensing signal image. The collected signals are evenly divided into data in 1S according to the sampling frequency during signal collection, and event time sequence data in 1S are converted into GAF images corresponding to different events in 1S through a Granami angular field. The size of the obtained event data set is adjusted and the event data set is divided into a training set, a verification set and a test set according to the proportion, and the proportion corresponding to the three data sets is as follows: 8:1:1.
And step 3: two different deep convolutional neural networks are designed, one is a teacher network model, and the other is a student network model. The teacher network model is pre-trained on the ImageNet dataset and the optimal training weights are saved. The student network model is pre-trained on the MNIST public data set and the best pre-selected weights are saved.
And 4, step 4: and (3) performing knowledge distillation training on the optimal training weight of the double migration networks, the teacher model and the student model in the step (3) on the event GAF image data set, and storing the training model parameters of the optimal student model.
And 5: and (4) carrying out unknown distributed optical fiber sensing event identification on the student model trained in the step (4).
Fig. 2 shows a structure and a working principle of a distributed fiber acoustic wave/vibration sensing system based on a phase-sensitive optical time domain reflectometry (Φ -OTDR) technique used in step 1. The system hardware for collecting signals comprises optical signal demodulation equipment, a signal processing host and a detection optical cable.
The optical signal demodulation device is a core device of a distributed optical fiber sensing system, as can be seen in fig. 2, a Narrow Line Laser (NLL) is used as a light source of laser, continuous Waves (CW) emitted by the narrow line laser are firstly modulated into corresponding optical pulse signals through an acousto-optic modulator (AOM), the optical pulse signals are amplified through a first erbium-doped amplifier (EDFA 1), filtered through an optical fiber bragg grating Filter (Filter), and then transmitted into a single-mode optical fiber through a circulator, and then backscattered through a second erbium-doped amplifier (EDFA 2) and the optical fiber bragg grating Filter to obtain better improvement of signal-to-noise ratio, and then injected into a 3*3 Coupler (Coupler), two ports on the other side of the Coupler are connected with two third rotating mirrors (FRM), incident light is divided into two beams, reflected by the faraday rotating mirrors and interfered in the Coupler, and the interfered signals are received by three Balanced Photodetectors (BPD), and finally, faraday data is collected and stored in a personal computer.
Dividing the distributed optical fiber sensing signal data into single-channel event data according to the optical fiber sensing positioning information and the sampling frequency information of the collected event signals, wherein the corresponding data style is as follows:
。
the collected event data are converted into two-dimensional event image data through a Gramami Angular Field (GAF), and the two-dimensional event image data are converted into event time domain correlation, and meanwhile, the GAF method is insensitive to power supply waves in a light path and can be better combined with a deep learning network.
The process of converting the GAF event timing signal data into a corresponding event GAF image is shown in fig. 3. Collecting the event sequence
Wherein->
The length of the time sequence is obtained by firstly mapping one-dimensional time sequence data to [ -1,1 through a Min-Max Scaler normalization method]The corresponding formula of the corresponding normalization method is as follows:
converting the time sequence processed by normalization into a polar coordinate, namely regarding the numerical value as a cosine value of an included angle, regarding the timestamp as a radius, and taking the corresponding formula as follows:
in equation (11), N is a constant used as a polar space regularization factor, and the time dependency can be maintained by converting the time series into polar coordinates.
After the numerical values are mapped to polar coordinates, a corresponding Gramami angular field matrix is obtained by calculating a trigonometric function value between two points, and the corresponding formula is as follows:
and (3) scaling the elements in the Granami angular field matrix in the formula (12) to 0-255, so as to obtain an event GAF two-dimensional event image.
And (3) preprocessing the obtained event image, uniformly scaling the obtained event image into a color RGB image with the size of 224 × 224, and dividing the ratio of 8.
Event data are converted into GAF image data sets of different events through a GAF method, and a large teacher network model and a light student network model are designed in the embodiment. In the industrial field and common convolutional neural network models, resNet50 is selected as a teacher network model, the ResNet50 is pre-trained on an ImageNet data set, and the optimal weight of the teacher network model is stored.
The design of the custom student network model is carried out, wherein an ICEblock module in the student network model is shown in figure 4 and mainly comprises an inverted residual error structure, CBAM space and channel attention memory and an ECANet attention mechanism. Unlike the residual structure in the Resnet50 teacher network, in the inverted residual structure of the student network model, the deep separable convolution is a form of deconvolution, which is mainly composed of channel-by-channel convolution and point-by-point convolution. One convolution kernel of the channel-by-channel convolution is only responsible for one channel, one channel is only convoluted by one convolution kernel, and the point-by-point convolution can perform weighted combination on the feature vector diagram of each step in the depth direction to form a new feature vector diagram. The common convolution is theoretically 8 to 9 times more computationally intensive than the depth separable convolution.
In ICEblock, as shown in FIG. 4, a1 × 1 convolution is used for data dimension promotion, a 3 × 3 convolution is used for depth residual error convolution, a CBAM module is used for merging a channel attention mechanism and a space attention mechanism, a1 × 1 convolution is used for data dimension reduction operation, short linking is used in a residual error structure only when the dimensions of input features and output features in the inverse residual error structure are the same, and ECANet is introduced into the inverse residual error structure for merging the attention mechanisms again.
The attention mechanism module CBAM is composed of a channel attention mechanism and a spatial attention mechanism, as shown in fig. 5, the attention mechanism can be regarded as a dynamic selection process of important information of image input, and the process is realized by feature adaptive weight. Specifically, firstly, the input feature map is input
Two different spatial semantic description operators are obtained through two parallel MaxPool layers and AvgPool layers: />
And &>
The two are respectively used as the input of a shared multilayer perceptron comprising a hidden layer to generate the channel attention feature vector. In order to reduce the parameter number, the number of the hidden layer neurons is C/r, wherein r is a channel reduction rate, elements corresponding to the obtained two channel attention characteristic vectors are added, and then an activation function is used to obtain the final channel attention diagram ^ based on the judgment result>
. The specific channel attention process may be represented by the following equation:
（13）
in the formula (I), the compound is shown in the specification,
for Sigmoid activation function, <' >>
，/>
Is the weight of the shared multi-tier perceptron. />
，/>
For the results of the parallel averaging pooling layer and the maximum pooling layer, the channel attention map->
Each weight in the feature map represents the importance and the degree of association for the key information in the feature map of the corresponding channel.
In a spatial attention-performing procedure, which utilizes features after being reconstructedThe spatial relationship of the graphs generates a spatial attention graph. The average pooling and the maximum pooling are simultaneously carried out along the channel direction to aggregate the channel information of the input feature map, and two-dimensional channel feature description operators are respectively obtained
And &>
Merging the two characteristics, carrying out dimension splicing to generate effective space matrix characteristics, obtaining a space attention matrix through 7 multiplied by 7 convolution, and obtaining a two-dimensional space attention diagram based on a sigmoid activation function>
It contains the spatial position of the information to be focused or suppressed, and the specific spatial attention calculation formula is:
in the formula (I), the compound is shown in the specification,
for Sigmoid activation function, <' >>
Denoted as 7 x 7 convolution operation. />
Represented as the output of the channel attention module, <' > or>
，/>
Expressed as average pooling and maximum pooling, respectively. The space and channel attention mechanism complement each other, and the regional characteristics containing key information can be effectively highlighted.
The CBAM structure in ICEblock is shown in fig. 5, and the channel attention mechanism is to compress the feature map in the spatial dimension to obtain a one-dimensional vector and then operate. When compression is performed in the spatial dimension, not only global average pooling but also global maximum pooling is performed. Pooling is performed to aggregate the spatial information of the feature maps to a shared network, compress the spatial dimensions of the input feature map, and merge element-by-element to generate channel attention. The spatial attention mechanism is to compress the channel, perform average pooling and maximum pooling respectively in the channel dimension, the average pooling is to take the average value on the channel, the number of times of extraction is the maximum value on the channel drawn by the maximum height times width, the number of times of extraction is the maximum value on the channel, and then merge and fuse the previously extracted features.
The ECANet structure of ICEblock is as shown in fig. 6, in a module, global average pooling operation is firstly performed on each channel to obtain 1 × 1 × C channel data, convolution operation is performed through 1D sliding window convolution with shared weights, the size of a 1D convolution kernel is K, the number of channels C increases in a power of 2 form as the network deepens, and the corresponding relation between K and C is as follows:
in the formula (15)
And &>
Is taken to be 2 and 1, is>
Expressed as an adjacent odd value of the calculated value. Cross-channel interaction and channel reduction are achieved through an adaptive convolution kernel K.
Specific parameters of the student network model are shown in table 1. In the student network model, the method comprises two parts, namely feature extraction and a classifier, wherein the feature extraction part comprises a convolution kernel of 3*3 with a step of 2 and Conv2d with a padding of 1, 10 proposed ICEblocks, and finally Conv2d with a convolution kernel of 1*1 with a step of 1. In the classifier part, firstly Dropout discards part of neural network units from the network according to the probability of 0.25, and then linear is used for feature splicing and mapping to the event classes.
TABLE 1 student network architecture
And pre-training the custom designed student network model on an MNIST public data set and storing the optimal pre-selected weights.
And training a model capable of realizing different event classifications by combining an event GAF image classification method of double-migration learning and knowledge distillation. And transferring the ResNet50 pre-trained on ImageNet and the model parameter matrix of the custom network pre-trained on MNIST to a GAF event data set to classify the problems by the optical fiber sensing events, and sequentially constructing a teacher model and a student model by the two models when the models are trained by the method. The detailed knowledge distillation model diagram is shown in FIG. 7. The specific distillation process is as follows:
Step 2: will migrate to trainLoading the optimal student network model, inputting event GAF image information into the model in the form of pixel matrix to obtain probability distribution of different events, and successively dividing by temperature
Smoothing is carried out, and then the soft label on the event data set is obtained by the softmax function>
Corresponding calculation formula (17), in which->
Is a feature map of the last layer of the student network model.
And step 3: loading the optimal student network model for transfer training, inputting event GAF image information into the model in the form of pixel matrix to obtain probability distribution of different events, and directly obtaining hard labels of the student network model by softmax function
Corresponding calculation formula (18).
And 4, step 4: calculating the soft label of the teacher network model obtained in the step 1 and the soft label of the student network model obtained in the step 2 through a relative entropy loss function to obtain a soft loss function
The corresponding calculation formula is (19), in formula (19), B is the number of pictures to be processed in batch, and C is the number of category types of the event GAF image.
And 5: calculating the hard label of the student network model obtained in the step 3 on the GAF event data set and the real label of the event GAF image by using a cross entropy loss function to obtain a corresponding hard loss function value
The calculation formula is (20):
step 6: the soft loss function obtained in the step 4 and the hard loss function obtained in the step 5 are subjected to coefficient of proportionality
Adjusting the ratio of the two loss values results in a mixing loss function->
The corresponding calculation formula is (20), will get->
The back propagation is the key for the teacher model to give the student model the dark knowledge and greatly improve the classification performance of the student model.
（21）
And 7: and performing distillation training on the teacher model and the student model of the double-migration network on the GAF event data set, and storing the optimal training result of the student network model.
Example 3
An embodiment of the present disclosure provides a distributed optical fiber sensing event recognition system based on a glatiramer angular field, including:
the data signal acquisition module is used for acquiring optical fiber sensing signals acquired in a set scene and position information of distributed optical fiber sensors acquiring the optical fiber sensing signals in the set scene;
the event data conversion module is used for extracting event data in the optical fiber sensing signals according to sampling frequency when the optical fiber sensing signals are collected and position information of the distributed optical fiber sensors; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
and the event recognition module is used for inputting the acquired event image data into the student model which is trained by double migration and knowledge distillation to acquire a recognition result.
The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Although the embodiments of the present disclosure have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present disclosure, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive changes in the technical solutions of the present disclosure.
Claims (10)
1. A distributed optical fiber sensing event identification method based on a Grarami angular field is characterized by comprising the following steps:
acquiring optical fiber sensing signals acquired in a set scene and position information of a distributed optical fiber sensor acquiring the optical fiber sensing signals in the set scene;
extracting event data in the optical fiber sensing signal according to the sampling frequency when the optical fiber sensing signal is acquired and the position information of the distributed optical fiber sensor; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
and inputting the acquired event image data into a student model trained by double migration and knowledge distillation to acquire a recognition result.
2. The method according to claim 1, wherein the collected optical fiber sensing signals are one-dimensional signal data.
3. The distributed optical fiber sensing event identification method based on the Granami angle field as claimed in claim 1, wherein the process of cutting the event data into time sequence data according to the sampling frequency during the optical fiber sensing signal acquisition and converting the time sequence data into corresponding event image data through the Granami angle field is to divide the time sequence data into data in 1S uniformly according to the sampling frequency during the optical fiber sensing signal acquisition, and convert the event time sequence data in 1S into corresponding different event Grara Mi Jiao field image data in 1S through the Granami angle field.
4. The distributed optical fiber sensing event recognition method based on the grassplot angular field of claim 1, wherein the step of obtaining the student model trained by the double migration and knowledge distillation comprises: designing two different deep convolutional neural networks which are respectively a teacher network model and a student network model, pre-training the teacher network model on a data set and storing the optimal training weight; the student network model is pre-trained on public data sets and the best pre-selected weights are saved.
5. The method of claim 4, wherein the optimal training weights of the dual migration networks and the teacher model and the student models are trained for knowledge distillation on the GAF image event data set, and the training model parameters of the optimal student models are saved as trained student models for identifying the unknown distribution as the fiber sensing event.
6. The distributed fiber sensing event identification method based on the Grarami angular field according to claim 4, characterized in that the main structure of the student network model is an inverse residual structure, and a space and channel attention CBAM module and a local cross-channel interaction strategy module ECANet are added into the inverse residual structure.
7. The distributed optical fiber sensing event identification method based on the Grarami angular field as claimed in claim 4, characterized in that the optimal teacher network for the migration training is loaded, GAF event image information is input into the model in the form of pixel matrix to obtain corresponding probability distribution of different events, smoothing is performed through successive division by temperature, then soft labels of the teacher on the event data set are obtained through softmax function, and soft loss function is obtained through calculation of the obtained soft labels of the teacher network model through relative entropy loss function.
8. Distributed optical fiber sensing event identification system based on the glatiramer angular field is characterized by comprising:
the data signal acquisition module is used for acquiring optical fiber sensing signals acquired in a set scene and position information of distributed optical fiber sensors acquiring the optical fiber sensing signals in the set scene;
the event data conversion module is used for extracting event data in the optical fiber sensing signals according to sampling frequency when the optical fiber sensing signals are collected and position information of the distributed optical fiber sensors; cutting event data into time sequence data according to distributed optical fiber sensing by adopting frequency, and converting the time sequence data into corresponding event image data through a Granami angular field;
and the event recognition module is used for inputting the acquired event image data into the student model which is trained by double migration and knowledge distillation to acquire a recognition result.
9. The grammide angular field based distributed optical fiber sensing event recognition system of claim 8, wherein the collected optical fiber sensing signals are one-dimensional signal data.
10. The grammide angular field based distributed optical fiber sensing event recognition system of claim 8, wherein the optimal training weights of the dual migration network and the teacher model and the student models perform knowledge distillation training on the event GAF image dataset and save the training model parameters of the optimal student models.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310125487.5A CN115859186B (en) | 2023-02-17 | 2023-02-17 | Distributed optical fiber sensing event identification method and system based on Grami angle field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310125487.5A CN115859186B (en) | 2023-02-17 | 2023-02-17 | Distributed optical fiber sensing event identification method and system based on Grami angle field |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115859186A true CN115859186A (en) | 2023-03-28 |
| CN115859186B CN115859186B (en) | 2023-06-02 |
Family
ID=85658287
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310125487.5A Active CN115859186B (en) | 2023-02-17 | 2023-02-17 | Distributed optical fiber sensing event identification method and system based on Grami angle field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115859186B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117312828A (en) * | 2023-09-28 | 2023-12-29 | 光谷技术有限公司 | Public facility monitoring method and system |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200251213A1 (en) * | 2016-05-02 | 2020-08-06 | Bao Tran | Blockchain gene system |
| CN114841202A (en) * | 2022-04-24 | 2022-08-02 | 浙江浙能天然气运行有限公司 | Deep learning mode identification method for distributed optical fiber pipeline intrusion detection |
| CN115115924A (en) * | 2022-07-20 | 2022-09-27 | 河海大学 | Concrete image crack type rapid intelligent identification method based on IR7-EC network |
| CN115294498A (en) * | 2022-08-05 | 2022-11-04 | 山东大学 | Cross attention-based multimode fusion wavelet knowledge distillation video behavior identification method and system |
| CN115457412A (en) * | 2022-08-31 | 2022-12-09 | 河海大学 | Fast concrete crack identification method based on Faster-rIR7-EC |
| CN115496100A (en) * | 2022-09-22 | 2022-12-20 | 南昌航空大学 | Perimeter security disturbance identification algorithm based on GAF-ConvNeXt-TF |
-
2023
- 2023-02-17 CN CN202310125487.5A patent/CN115859186B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200251213A1 (en) * | 2016-05-02 | 2020-08-06 | Bao Tran | Blockchain gene system |
| CN114841202A (en) * | 2022-04-24 | 2022-08-02 | 浙江浙能天然气运行有限公司 | Deep learning mode identification method for distributed optical fiber pipeline intrusion detection |
| CN115115924A (en) * | 2022-07-20 | 2022-09-27 | 河海大学 | Concrete image crack type rapid intelligent identification method based on IR7-EC network |
| CN115294498A (en) * | 2022-08-05 | 2022-11-04 | 山东大学 | Cross attention-based multimode fusion wavelet knowledge distillation video behavior identification method and system |
| CN115457412A (en) * | 2022-08-31 | 2022-12-09 | 河海大学 | Fast concrete crack identification method based on Faster-rIR7-EC |
| CN115496100A (en) * | 2022-09-22 | 2022-12-20 | 南昌航空大学 | Perimeter security disturbance identification algorithm based on GAF-ConvNeXt-TF |
Non-Patent Citations (3)
| Title |
|---|
| ZHU CY ET AL: "《A Comprehensive Bibliometric Analysis of Signal Processing and Pattern Recognition based on Distributed Optical Fiber 》", 《SCIENCEDIRECT》 * |
| 刘平: "《对地观测中基于迁移学习的虚警源检测与识别技术》", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 王春晓: "《基于长短期记忆神经网络的光性能监测技术》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117312828A (en) * | 2023-09-28 | 2023-12-29 | 光谷技术有限公司 | Public facility monitoring method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115859186B (en) | 2023-06-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112949565B (en) | Single-sample partially-shielded face recognition method and system based on attention mechanism | |
| CN111696196B (en) | Three-dimensional face model reconstruction method and device | |
| CN111563562B (en) | Color target reconstruction method of single-frame scattering image based on convolutional neural network | |
| CN115859186B (en) | Distributed optical fiber sensing event identification method and system based on Grami angle field | |
| CN111598028A (en) | Method for identifying earth surface vegetation distribution based on remote sensing imaging principle | |
| CN116385902A (en) | Remote sensing big data processing method, system and cloud platform | |
| Zhang et al. | Deep unfolding for singular value decomposition compressed ghost imaging | |
| CN116824629A (en) | High-robustness gesture recognition method based on millimeter wave radar | |
| Zhou et al. | MSAR‐DefogNet: Lightweight cloud removal network for high resolution remote sensing images based on multi scale convolution | |
| CN117132759A (en) | Saliency target detection method based on multiband visual image perception and fusion | |
| Ran et al. | RGAN: Rethinking generative adversarial networks for cloud removal | |
| CN110942097A (en) | Imaging-free classification method and system based on single-pixel detector | |
| Shit et al. | An encoder‐decoder based CNN architecture using end to end dehaze and detection network for proper image visualization and detection | |
| CN114202473A (en) | Image restoration method and device based on multi-scale features and attention mechanism | |
| Chen et al. | Attentive generative adversarial network for removing thin cloud from a single remote sensing image | |
| CN116091640B (en) | Remote sensing hyperspectral reconstruction method and system based on spectrum self-attention mechanism | |
| CN116453277A (en) | Perimeter security intrusion event full coverage accurate detection method and system | |
| CN116091897A (en) | Distributed optical fiber sensing event identification method and system based on light weight | |
| CN115496788A (en) | Deep completion method using airspace propagation post-processing module | |
| CN115331162A (en) | Cross-scale infrared pedestrian detection method, system, medium, equipment and terminal | |
| CN115035408A (en) | Unmanned aerial vehicle image tree species classification method based on transfer learning and attention mechanism | |
| Zhao et al. | Robust single-photon 3D imaging based on full-scale feature integration and intensity edge guidance | |
| Gao et al. | CP-Net: Channel attention and pixel attention network for single image dehazing | |
| CN116504005B (en) | Perimeter security intrusion signal identification method based on improved CDIL-Bi-LSTM | |
| Yang et al. | Improving Semantic Segmentation Performance by Jointly Using High Resolution Remote Sensing Image and Ndsm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |